E. coli BL21 were purchased from Invitrogen. Plasmid pET-21A was purchased from Novagen. pDNR-LIR-hVDAC was purchased from Dharmacon. Ampicillin, IPTG, Triton 100X, Guanidine hydrochloride, Acrylamide: Bisacrylamide 37.5:1, SDS 10%, Temed, APS, TGS 10X, NH₄Cl and CaCl₂ were purchased from Euromedex. NaCl, NaH₂PO₄, K₂HPO₄, KH₂PO₄, glucose, LB-broth, Tris-HCl, NADH, propofol, quinidine, ubiquinone 0, Laemli 6X, β-mercaptoethanol, PBS, Na₂HPO₄, K₂SO₄ and metformin were purchased from ThermoFisher. Aspirin, catechin, curcumin, DCCD, DPC, emodin, fluoxetine, and MnCl₂ were purchased from Fluorochem. Imidazole, DIDS, DMSO, ReadyBlue Protein Gel stain, EDTA, MgCl₂, CoCl₂, H₂BO₃, acetone and ethanol were purchased from Sigma-Aldrich. FeSO₄ and ZnSo₄ were purchased from Biobasic. CuCl₂ and Mo₇O₂₄ were purchased from Roth. ¹⁵NH₄Cl, D₂O and deuterated glucose were purchased from Innova-Chem. LDAO was purchased from CliniSciences. Itraconazole was purchased from Cayman Chemical company. VBIT4 was purchased from Aobious. Cannabidiol was purchased from Carbosynth. Olesoxime was obtained from in-house chemical library.

hVDAC1 cDNA was subcloned into pET21A from pDNR-LIR-hVDAC. hVDAC1 was produced in BL21 E. coli transformed with a pET-21a plasmid (supplemental Figure S1A) containing the hVDAC1-coding sequence and purified as described in³⁹. The protein was extracted from inclusion bodies in Guanidium containing buffer. The denatured protein was purified by affinity chromatography on a HisTalon superflow Cartridge column (Takara bio-Company) followed by size-exclusion chromatography (SEC; HiLoad 16/600 Superdex 200 pg, Cytiva). VDAC1 refolding was performed by dropwise dilution into a buffer supplemented with 1% LDAO³⁹. The protein was purified by cation exchange chromatography (HiLoad 16/10 SP Sepharose HP; Cytiva) followed by a SEC. Protein purity was verified by SDS-PAGE (supplemental Figure S1B) and Coomassie staining. Proper folding of the protein was confirmed by ¹H-¹⁵N TROSY-HSQC performed with 90 µM of perdeuterated hVDAC1 in a 5 mm tube on a 950 MHz Bruker spectrometer equipped with a cryogenically cooled probe (supplemental Figure S2).

hVDAC1 (50 µg/mL) was incubated 20 min with each compound in the experimental buffer (PBS, pH 7.2 with 0.1% LDAO) containing either 0% or 5% DMSO depending on compound solubility. Next, samples were loaded into NanoDSF Standard-grade capillaries and transferred into a Prometheus NT.48 NanoDSF device (NanoTemper Technologies). Thermal unfolding was detected during heating in a linear thermal ramp (2 °C/min; 30–80 °C) with an excitation power of 20%. Unfolding transition points were determined from changes in the emission wavelengths of tryptophan fluorescence at 330 nm and 350 nm, respectively, and from the ratio 350/330 nm. Experiments were conducted in triplicates and data were analysed using the Prometheus PR stability analysis software.

Purified hVDAC1 was fluorescently labelled using NanoTemper Protein labelling kit RED-NHS 2nd generation (L011, NanoTemper Technologies) or labelling kit BLUE-NHS (L003) depending on ligand fluorescence according to manufacturer instructions. hVDAC1 (100 nM) was incubated 20 min with each compound in the dark at room temperature at the highest soluble ligand concentration (binding check experiment) or at 16 different concentrations obtained by serial dilution starting from the highest soluble concentration (dose-response experiment). Samples were loaded into glass capillaries (Monolith NT Capillaries), and thermophoresis analysis was performed using a Nano-Temper Monolith NT.115 apparatus. (LED power 20%, IR laser power 20%). Signal quality was monitored by the NanoTemper Monolith device to detect possible ligand auto-fluorescence, precipitation, aggregation, or ligand-induced changes in the photobleaching rate. Experiments were conducted as triplicates and treated with the MO.Affinity Analysis software (Nano-Temper).

About 13 ligands of VDAC, excluding analogues, have been described in the literature so far. One of the most recent and complete review on VDAC ligands⁴⁰ was used for the selection of the compounds tested in the present work. A high structural heterogeneity in the selected set was a primordial criterion. Fragment-size molecules (aspirin, propofol) to much more complex molecules (itraconazole) were picked by keeping, at most, two compounds of the same family. The selected molecules also included ligands expected to bind inside the pore region (aspirin, curcumin, emodin…)^33,41 and ligands most likely binding to the outer, i.e., lipid-facing, part of the channel (propofol, olesoxime)^42,43. Importantly, DCCD was shown to covalently bind to VDAC⁴⁴. As reported for other channels, DIDS might as well bind covalently to VDAC⁴⁵. The seventeen compounds investigated in the present study are listed in Table 1 where we also specified the original experimental technique(s) used to demonstrate direct interaction and the reported effect(s) on channel activity.

Recombinant human VDAC1 (hVDAC1) samples were produced according to the protocol published by Silver et Eddy³⁹. The proper refolding of the protein was assessed by ¹H-¹⁵N TROSY-HSQC NMR experiment. The NMR spectrum (supplemental Figure S2) was comparable to those published^14,15, which confirmed the correct refolding of the protein.

NanoDSF is a label-free thermal-shift-assay (TSA) technique used to study protein denaturation as a function of the temperature. NanoDSF provides indications on protein stability and enables to determine the binding ability of small molecules for targets of interest. The technique is based on the estimation of the melting temperature (T_m) of the protein by monitoring the intrinsic fluorescence of tryptophan residues. When excited at wavelengths around 280 nm, Tryptophans emit fluorescence as a peak around 330 nm when they are embedded in hydrophobic areas of the folded state, while they emit at 350 nm when they become exposed to the hydrophilic solvent during unfolding⁵⁸. hVDAC1 structure includes four tryptophans (W64, W75, W149 and W210) facing the lipid membrane. Varying the fluorescence intensity ratio (F350 nm/F330 nm) as a function of the temperature resulted in a S-shaped transition from the folded to the denatured state, usually referred to as redshift, which is the most common transition observed by NanoDSF⁵⁹. The melting temperature was first evaluated on hVDAC1 alone in the absence of ligands (Figure 2). Average T_m values were determined at 57.6 ± 0.05 °C and 56 ± 0.07 °C, at 0% and 5% DMSO, respectively, which is comparable to the T_m measured by TSA on CPM-labelled mVDAC1, i.e., T_m = 55.8 ± 0.6 °C in the absence of DMSO⁶⁰. Next, T_m values were determined in the presence of the seventeen selected compounds and compared to the T_m of hVDAC1 alone at 0%, or 5% DMSO when the compound was not soluble at 0% DMSO. The thermal shift (ΔT_m) was considered significant when it exceeded three times the standard deviation of the T_m in the absence of a ligand (± 0.22 °C). In this case, the compound was assumed to be a genuine ligand of hVDAC1.

As shown in Figure 3, 13 out of the seventeen compounds have been confirmed as VDAC1 ligands as they induced a significant shift in the T_m. Among them, six molecules (aspirin, cannabidiol, catechin, DCCD, fluoxetine and propofol) were found to increase the T_m. In contrast, seven molecules (curcumin, DIDS, DPC, olesoxime, quinidine, ubiquinone, VBIT4) induced a negative T_m shift. The four remaining compounds (emodin, itraconazole, NADH and metformin) have not been confirmed as VDAC1 ligands by NanoDSF in our experimental conditions.

MST enables to track temperature-induced changes in the fluorescence of a fluorochrome-tagged protein. When a ligand binds to a target, its MST signal is modified in a ligand concentration-dependent manner⁶¹. Therefore, the MST technique can either be used as a single-dose assay (i.e., at fixed ligand concentration) to determine the binding ability of a compound or at multiple concentrations in order to build dose-response curves from which the dissociation constant can be estimated. In this study, both approaches were applied. The seventeen compounds were first tested by single-dose assay referred to as binding check (Figure 4). A significant modification in the MST signal compared to the reference signal (protein alone) was observed for every compound except for ubiquinone. No information about the binding ability of catechin was obtained as auto-fluorescence, photobleaching rate modification and aggregation of the protein were systematically observed upon addition of this compound.

The fifteen ligands confirmed by single-dose assay were further investigated to determine their dissociation constant (K_d). Transition of the signal from the saturated-protein state (high ligand concentration) to the free-protein state (low ligand concentration) was monitored to determine the K_d (Figure 5). Overall, dissociation constants were successfully determined for six compounds including DIDS (K_d = 0.5 µM), VBIT4 (K_d = 3 µM), itraconazole (K_d = 5 µM), cannabidiol (K_d = 6 µM), curcumin (K_d = 6 µM) and emodin (K_d = 10 µM). As mentioned above, DIDS is likely to interact covalently with VDAC1. Therefore, the measured K_d must be carefully considered. Evaluation of the K_d was not possible for propofol and fluoxetine since the saturation plateau could not be reached (K_d > 650 µM for propofol and K_d > 7.7 mM for fluoxetine). The dissociation constant of six compounds (aspirin, DPC, metformin, NADH, olesoxime, and quinidine) could not be determined despite being validated by binding check. Indeed, no transition from the unbound to the bound state was observed for these ligands, which was likely due to a poor ligand affinity or a too weak modification of the fluorescence resulting in a too small signal amplitude. In contrast with a previous study⁴⁹, no proper dose-response data were obtained for DPC preventing the evaluation of the K_d. Similarly, we could not obtain dose-response data for DCCD which, in this case, may be related to the ability of the compound to form covalent bonds with proteins^44,46.

K_d values determined for cannabidiol (6 µM) and VBIT4 (3 µM) were in good agreement with published data, i.e., 11 µM for cannabidiol³⁴ and 53 µM for VBIT4²⁹, while a significant difference was found for itraconazole (K_d = 5 vs 120 µM in²⁶). Finally, the present work is reporting for the first time the affinity of curcumin and emodin for VDAC1, with K_d values of 6 µM and 10 µM, respectively.

A summary of the results obtained with both techniques (NanoDSF and MST) is provided in Table 2. Every tested compound was confirmed as ligand by at least one technique and eleven out of seventeen were confirmed with both single dose assays. Four molecules did not induce a significant T_m shift in NanoDSF but were confirmed as ligands by MST binding check. Overall, only two ligands were not confirmed by binding check (ubiquinone and catechin) whereas they induced a significant T_m shift in NanoDSF. As no ligands were simultaneously unvalidated by both methods, this suggests that the combination of NanoDSF and MST might provide an efficient way to reduce false negatives in low- or medium-throughput screening campaigns.